An Intelligent Hybrid Machine Learning Model for Paddy Disease Detection

Abstract

Most countries rely on paddy/rice as a preferred staple crop due to its desirable agronomic characteristics. However, during the growth cycle, undetected or delayed identification of diseases can significantly reduce crop yield. To address this, machine learning approaches are increasingly employed for early-stage disease detection. Traditional Recurrent Neural Networks (RNNs), while useful for sequential data, suffer from limitations such as inadequate memory retention, architectural complexity, and slower processing speeds. To overcome these challenges, this study proposes a hybrid deep learning model integrating RNN, Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). In this framework, the RNN component captures temporal dynamics by accommodating variable intervals, such as daily, weekly, or custom-defined gaps, between image inputs for time-series analysis. The LSTM component ensures long-term sequential memory retention, while the GRU contributes a streamlined architecture that accelerates processing. The proposed model enables early disease identification and provides actionable recommendations, such as initiating pesticide application or determining the optimal time for harvest. It operates iteratively, reassessing disease status, whether eliminated or persistent, based on initial detection and subsequent input intervals. By integrating data preprocessing techniques and a well-defined predictive structure, the model achieves near-optimal accuracy and performance, thereby minimizing crop damage and enhancing overall yield.